AlloyDB for PostgreSQL release notes

You can now use the Model Context Protocol (MCP) Toolbox for Databases to access AlloyDB observability features and advanced query insights directly in your IDE.

You can now sync tables from BigQuery into your AlloyDB instance, either as a one-time operation or on a periodic schedule. This feature (in Preview) lets you enable operational analytics that benefit from low-latency, transactional access to your data lake.

For more information, see Sync BigQuery data to AlloyDB.

AlloyDB integration with BigQuery lets you connect your operational and analytical data through real-time data access (lakehouse federation), periodic data synchronization, and one-time table syncs. These features are in Preview.

For more information, see Choose how to access BigQuery data from AlloyDB.

AlloyDB now supports Best Matching 25 (BM25) indexes for full-text search in Preview. You can use the pg_textsearch extension to create BM25 indexes and optimize probabilistic ranking of full-text search. This feature is supported on AlloyDB instances running PostgreSQL 17 or 18.

For more information, see Create and manage a BM25 index.

IAM group authentication for AlloyDB is available in Preview for new clusters running PostgreSQL 15 and later. This feature simplifies database user management by allowing access management at the group level, where group members inherit database roles and permissions. To use this feature, enable the alloydb.iam_authentication and alloydb.iam_group_authentication database flags.

For more information, see IAM group authentication and Manage IAM authentication.

AlloyDB write endpoints are now available in Preview. Write endpoints simplify database connection management by providing a stable domain name service (DNS) name for your applications, decoupling them from instance IP addresses. During disaster recovery switchovers or failovers, AlloyDB automatically updates the endpoint to point to the new primary instance.

For more information, see Manage database connections with write endpoints.

AlloyDB now supports cross-region failover in Preview. You can optionally enable cross-region failover to automate the recreation of the original primary when a secondary cluster is promoted while maintaining your replication topology without requiring manual cluster deletion or re-creation. Cross-region failover is supported only for topologies with a single primary cluster and a single secondary cluster.

For more information, see About cross-region replication and Work with cross-region replication.

External search with AlloyDB now supports OpenSearch in Preview. You can use the external_search_fdw extension to connect to an OpenSearch cluster and query its data directly from your database.

Transparent query forwarding is now available in Preview for clusters compatible with PostgreSQL 17 and 18. With this feature, the primary node in a cluster intercepts read-only queries and selectively forwards them to read pool instances while maintaining read-your-writes consistency.

For more information, see Optimize resources and isolate read queries with transparent query forwarding.

External search with AlloyDB now supports Apache Solr in Preview. You can use the external_search_fdw extension to connect to a Solr cluster and query its data directly from your database.

AlloyDB integration with Knowledge Catalog is now generally available (GA).

This integration provides a unified metadata view to simplify data governance and analysis. It includes near real-time synchronization and expanded metadata details, like primary and foreign keys.

For more information, see Integrate AlloyDB with Knowledge Catalog.

The Database Insights remote Model Context Protocol (MCP) server now supports the following advanced query insights tools for AlloyDB for PostgreSQL:

• get_advanced_aggregated_query_stats
• get_advanced_aggregated_wait_event_stats
• get_advanced_time_series_query_stats
• get_advanced_time_series_wait_event_stats
• get_index_recommendations

For more information, see Database Insights remote MCP server.

You can publish an AlloyDB for PostgreSQL conversational analytics agent in Gemini Enterprise. This feature is in Preview.

You can now configure a cooldown period to determine when autoscaling occurs for your read pool instances. For more information, see Scale an instance.

ChatGPT users are now able to list and use the AlloyDB toolset provided by the AlloyDB remote MCP server.

AlloyDB now offers extended support for clusters running major PostgreSQL versions that have reached their end-of-life (EOL) as defined by the PostgreSQL community. Extended support provides an additional three years of support after the end of regular support, giving you more time to plan and perform major version upgrades. For more information, see Extended support for AlloyDB for PostgreSQL.

When the initial user or password is unspecified during cluster creation, a locked postgres role with null password is created.

Database Migration Service quick-start migrations (in Preview) are now integrated into AlloyDB to provide a lightweight, continuous migration flow. This feature automates setup for sources with private IPs in a VPC network, including Cloud SQL for PostgreSQL and self-managed databases on Compute Engine.

For more information, see Quick-start migrations overview in the Database Migration Service documentation.

The integration between AlloyDB for PostgreSQL and Knowledge Catalog (in Preview) is now enabled by default for all new AlloyDB clusters.

This integration provides a unified view of your metadata to simplify data governance and analysis. Recent enhancements feature near real-time synchronization, with updates reflecting in Knowledge Catalog within minutes, as well as expanded metadata details that now include Primary Keys and Foreign Keys.

For more information, see Integrate AlloyDB for PostgreSQL with Knowledge Catalog.

You can now use the AlloyDB columnar engine to act as a read-optimized, in-memory cache for HNSW indexes. This increases the number of queries per second (QPS) that your database can handle for vector search workloads. This feature is available in Preview.

For more information, see Accelerate vector search with the columnar engine.

You can now use the Database Insights remote MCP server to analyze AlloyDB's performance and system metrics. This feature is in GA.

The AlloyDB remote MCP server—available through the global endpoint—is now generally available (GA) and includes support for the following:

• Read-only execute SQL.
• Tools to help you update your instance, export and import data, create a backup, and restore your cluster.
• Public IP.
• Private Service Connect endpoint automation.

For more information, see Use the AlloyDB remote MCP server.

ChatGPT users aren't able to list or use the AlloyDB toolset provided by the AlloyDB remote MCP server.

When querying your Elasticsearch data using standard SQL queries and specifying an OFFSET, if the OFFSET gets pushed down, it gets applied twice. For example, if your SQL query contains OFFSET 5, AlloyDB tries to push the OFFSET down. Then, AlloyDB applies the OFFSET again when the results are returned.

External search with AlloyDB now supports Elasticsearch in Preview.

With this update, you can use the external_search_fdw extension to connect to Elasticsearch and perform hybrid searches within AlloyDB. This integration allows you to combine the capabilities of AlloyDB with Elasticsearch for advanced search scenarios. For more information, see Access Elasticsearch data from AlloyDB.

The following AlloyDB AI function capabilities are available in Preview:

• You can now use AI function acceleration and the new AI Function Apply node to run faster queries with AI functions. This feature optimizes the execution of SQL queries that use the ai.if and ai.rank functions in PostgreSQL 17. For more information, see Accelerate performance for queries with AI functions.
• You can now use optimized AI functions to accelerate your AI queries while reducing operational costs. By training a smaller, faster proxy model on a sample of your data, AlloyDB can process most AI queries locally and only fall back to a remote LLM when necessary. For more information, see Accelerate queries using optimized functions.

• You can now use the sentiment analysis and summarization functions. These functions let you process and analyze unstructured data directly in your database:

  • ai.analyze_sentiment: classifies the emotional tone of text as positive, negative, or neutral, helping you analyze real-time customer feedback from thousands of raw, unstructured product reviews.
  • ai.summarize: condenses lengthy text into its essential information. Use this to extract key decisions and action items from sources like meeting transcripts or technical documentation.
  • ai.agg_summarize: an aggregate function that processes multiple rows in a column to generate a single, unified summary for a group. For instance, you can summarize all reviews for a specific seller using a GROUP BY clause.

  For more information, see Evaluate sentiment and Summarize content.

The following vector search improvements are now available in Preview:

• AlloyDB now supports Vector assist. Vector assist is an AlloyDB extension that simplifies the deployment and management of your AlloyDB vector workloads. It helps you set up production-ready vector search capabilities, such as embedding generation, query optimization, and index creation for vector types like HNSW. For more information about vector assist, how it works, and its limitations, see Vector assist overview.
• You can now defer ScaNN index creation on an empty table or a table with insufficient rows until the table has sufficient data. For more information, see Create a ScaNN index.
• The alloydb_scann extension now supports four-level tree indexes, providing support for tables with up to 10 billion vector rows. For more information, see Four-level ScaNN tree indexes.

Adaptive filtering from inline filtering to pre-filtering is now generally available (GA). With AlloyDB AI, you can use adaptive filtering to optimize filtered vector searches. This feature enables the query optimizer to use cost-based analysis to dynamically choose the most efficient filtering strategy—either inline filtering or pre-filtering—based on real-time data distributions. This improves filtered vector search performance without requiring manual tuning or intervention. Note that the feature adaptive filtering from pre-filtering to inline filtering is still in Preview.

For more information, see Understand adaptive filtering in AlloyDB AI.

The alloydb_scann extension is updated to include the following vector search improvements. These features are generally available (GA):

• By default, new ScaNN vector index builds are automatically tuned. Manually-tuned indexes can be converted to automatically-tuned indexes. For more information, see Create a ScaNN index.
• You can now automatically maintain your ScaNN vector indexes. AlloyDB incrementally manages your index such that when your dataset grows, AlloyDB updates centroids and splits large outlier partitions to provide better QPS and search results. For more information, see Maintain indexes automatically.

The AlloyDB for PostgreSQL index advisor is enabled by default. It provides vector search index recommendations for Scalable Nearest Neighbors (ScaNN) indexes. For more information, see Use the index advisor.

As of April 10, 2026, you can create, run, and edit Gemini Cloud Assist investigations only if you have a Premium Support contract. You can use Gemini Cloud Assist investigations to monitor and troubleshoot your AlloyDB for PostgreSQL instance with AI assistance.

If you ran an investigation prior to April 10, 2026, then the results of the investigation continue to be available to you in the Google Cloud console.

The QueryData tool lets you to query the data in your database using conversational language and build data agents. For more information, see QueryData tool overview. This feature is available in (Preview).

The preview release increases the accuracy of SQL generation with value search queries which match values and their context within a database. Value search queries trigger automatically. QueryData also adds support for Parameterized secure views (PSVs) to help secure applications that use natural language queries. For more information, see Secure and control access to application data using parameterized secure views.

The gcloud beta alloydb connect command is now available in Preview. This command provides a simplified way to connect securely to AlloyDB instances by using the AlloyDB Auth Proxy and psql. For more information, see Connect using gcloud CLI.

You can now enable Advanced Query Insights on primary clusters which have secondary clusters configured. Advanced Query Insights is not supported on secondary clusters. If you perform a switchover, you must re-enable Advanced Query Insights on the new primary cluster.

Hot standby enhances the AlloyDB high availability (HA) architecture to improve failover times and to ensure consistent performance after failover. AlloyDB continuously replicates transactions to the standby node to keep caches warm and to ensure that the node is ready to take over quickly during a failover. This feature is generally available (GA) in PostgreSQL 18 and is automatically enabled for all new instances. For more information, see the AlloyDB high availability overview.

AlloyDB now offers conversational analytics, which lets users query their operational data using natural language. This feature is powered by the Conversational Analytics API, which can help you translate complex human dialog into precise database queries to provide actionable insights. This feature is in Preview. For more information, see Conversational analytics for AlloyDB overview.

Database server compatibility with PostgreSQL version 18 is now generally available (GA):

• You can create AlloyDB clusters with PostgreSQL 18 compatibility.
• You can upgrade existing AlloyDB clusters running PostgreSQL major versions 14, 15, 16, or 17 to PostgreSQL major version 18 with one click.
• You can use Database Migration Service to migrate databases to AlloyDB.

The following AlloyDB AI features are available in Preview:

• You can now use the ai.hybrid_search() function, which fuses results from each search type into a single list using the Reciprocal Rank Fusion (RRF) algorithm. For more information, see Run hybrid vector similarity search.

• AlloyDB supports the rum extension for complex full-text search operations. The rum extension extends standard GIN indexes by storing positional information directly in the index. This enables faster phrase searches and relevance ranking without needing to access the table data. For more information, see Create and manage a RUM index.

When no major version is specified, AlloyDB for PostgreSQL now defaults to PostgreSQL major version 17 for new clusters.

AlloyDB lets you monitor node-level metrics in Google Cloud console and Metrics Explorer to provide detailed troubleshooting guidance for read pools and to identify nodes causing performance regressions. For more information, see System insights metrics reference.

AlloyDB now supports the 2 vCPU C4A machine type (c4a-highmem-2-lssd), which is powered by Google Axion, Google's custom Arm-based processor. This expansion provides a smaller entry point and more flexibility for scaling your production workloads using Axion-based instances. For more information, see Choose an AlloyDB machine type.

AlloyDB enhanced backups are generally available (GA). You can now select the Enhanced tier during cluster creation, manage your project-level backups with tiered tabs, and delete an enhanced backup. For more information, see Manage enhanced backups.

AlloyDB enhanced backups are generally available (GA). You can now select the Enhanced tier during cluster creation, manage your project-level backups with tiered tabs, and delete an enhanced backup. For more information, see Manage enhanced backups.

The following AlloyDB AI features are now generally available (GA):

• Auto vector embeddings provide a scalable, automated solution for managing the lifecycle of vector embeddings for large-scale datasets, eliminating the need for manual reindexing or custom scripts. This feature keeps embeddings in sync with transactional data and now supports incremental refresh in manual mode, ensuring that embeddings are only generated for new or updated rows. Additionally, you can perform incremental table refreshes or migration up to 130x faster than traditional row-by-row processing using bulk mode, improving efficiency for semantic search and Retrieval Augmented Generation (RAG).

• AI functions integrate LLMs like Gemini to bring 'world knowledge' to your AlloyDB data and incorporate advanced semantic search and ranking capabilities directly into your SQL workflows. This feature includes out-of-the-box functions for filtering (ai.if), semantic ranking (ai.rank), generation (ai.generate), and forecasting (ai.forecast).

• Experience higher performance in AlloyDB AI by utilizing array-based AI functions. You can perform batch processing of natural language prompts directly within your SQL queries, significantly improving efficiency for large-scale AI operations. For more information, see Perform intelligent SQL queries using AI functions.

Gemini Cloud Assist investigation capabilities are now supported in AlloyDB (Preview).

For more information, see Troubleshoot slow queries with AI assistance.

AlloyDB now integrates with Database Center to provide prioritized health monitoring in the Google Cloud console. This integration highlights critical and high-priority risks, offering one-click navigation to recommended fixes and system insights for quick resolution. For more information, see Monitor the health of your AlloyDB clusters and instances.

AlloyDB performance snapshot and reports now support read pool instance nodes, providing deeper observability into read operations and replica-specific performance issues.

New best practices are available for securing generative AI agents using Model Context Protocol (MCP) with Google Cloud databases. This guide covers key security measures like least privilege, native database controls, and secure agent design to help you build safer AI applications. For more information, see Best practices for securing agent interactions with Model Context Protocol.

You can now make AI function calls in bulk rather than row-by-row, which lets you scale your intelligent workflows faster with new support for array-based processing. For more information, see Perform intelligent SQL queries using AI functions.

This feature is in Preview.

You can now use the AlloyDB remote MCP server. The AlloyDB remote MCP server lets you interact easily with AlloyDB clusters from LLMs, AI applications, and AI-enabled development platforms.

This feature is in Preview.

We are announcing the release of support for the AlloyDB language connectors and Auth Proxy with Auto IAM Authentication and managed connection pooling. This feature and the fix for the issue from below is available starting with maintenance version 20260107.02_05. Clusters with a maintenance window that may not have received this release can use self-service maintenance to perform a maintenance update.

Virtual columns for expressions is a feature of the columnar engine in Preview that significantly improves query performance and reduces CPU consumption. It caches the results of frequently used expressions, which is especially beneficial for analytical workloads on large datasets. This feature is supported for PostgreSQL version 16 and higher.

Database server compatibility with PostgreSQL version 18 is now available for preview (Preview). You can create AlloyDB clusters with PostgreSQL 18 compatibility.

Automatic IAM authentication is unavailable when you use managed connection pooling with the AlloyDB Auth Proxy and Language Connectors. To sign into your database without a password, use manual IAM authentication. For more information, see Connect using an IAM account

You can create AlloyDB cluster instances in Bangkok, Thailand (asia-southeast3). For more information, see AlloyDB locations and AlloyDB for PostgreSQL pricing.

AlloyDB now supports the Z3 machine series, which are powered by 4th generation Intel x86 processors (Sapphire Rapids) with Titanium SSD. These instances offer machine sizes, with up to 88 vCPU and 704 GiB RAM, that let you run storage-intensive workloads with large working datasets. For more information, see Choose an AlloyDB machine type. This feature is generally available (GA).

Memory usage estimation is more accurate for high-dimensional vector indexes. This fix prevents out of memory (OOM) errors by enforcing defined memory constraints throughout the index build process. You might need to increase your maintenance_work_mem settings to align with the real usage estimates.

The extension vector, which includes pgvector functions and operators, is updated to version 0.8.1.

Managed connection pooling is now generally available (GA). This feature optimizes resource usage to improve workload scalability and reliability. It is compatible with the AlloyDB Auth Proxy and Language Connectors. For more information, see Configure managed connection pooling.

AlloyDB database performance snapshot reports now include a SQL Report section, which lists the top 50 queries by total elapsed time, read I/O, and standard deviation of elapsed time. This helps you identify and optimize resource-intensive queries.

You can build data agents that interact with the data in your database using conversational language. Use these data agents as tools to empower your applications. For more information, see Data agents overview. This feature is available in Preview, and access to it requires a sign-up.

You can now use Gemini 3.0 Flash (Preview) when you call generative AI functions in AlloyDB, such as AI.GENERATE. Use the model name gemini-3-flash-preview. For more information, see Use Gemini 3.0 models.

AlloyDB now supports the C4 machine series, which are powered by 6th generation Intel Xeon Granite Rapids processors. These instances offer massive machine sizes, with up to 288 vCPU and 2232 GiB RAM, that let you run extremely demanding workloads. For more information, see Choose an AlloyDB machine type. This feature is generally available (GA).

You can now use Gemini to fix query errors in the AlloyDB Studio query editor. This feature is available in Preview.

Query plan management (Preview) ensures query plan stability, and protects your database performance against the risk of query plan regression due to changes in the database or the optimizer's behavior. AlloyDB continuously monitors, captures, and logs potential query execution plans, giving you the granular control to force the optimizer to choose from approved plans, and prevent unintended regressions. For more information, see Manage query plans.

AlloyDB now supports horizontal autoscaling for read pool instances. This feature is available in Preview.

You can now perform self-service maintenance if you need to apply the latest AlloyDB updates to your clusters as soon as possible. Updating to the latest version can unlock AlloyDB features, apply patches, and let you set deny periods.

The upper limit of the query plans captured per minute is enhanced to 200. For more information, see Improve query performance using advanced query insights features for AlloyDB.

You can now use Gemini 3.0 when you call generative AI functions in AlloyDB, such as AI.GENERATE. For more information, see Use Gemini 3.0 models.

AlloyDB AI native vector search accelerator is now generally available (GA). It includes the following features and improvements:

• The google_columnar_engine extension is updated to automatically recommend data for searches, so you don't need to manually add vector columns to the table. For more information, see Perform a vector search.
• The alloydb_scann extension is updated with new metrics for vector index creation. You can now use the pg_stat_ann_index_creation view to see the number of rows present in a table at the time of index creation.

Improve query performance for partitioned tables and complex queries using parallel execution for the SELECT portion of an INSERT INTO ... SELECT query. For more information, see Improve select insert performance with parallel select.

You can now use Dataplex Universal Catalog to discover, search, and manage your AlloyDB resources, including clusters, databases, and tables. This integration (in Preview) provides a unified view of your metadata, enabling better data governance and easier analysis. For more information, see Manage your AlloyDB resources using Dataplex Universal Catalog.

AlloyDB now supports PostgreSQL 17 for migrating from Cloud SQL for PostgreSQL to AlloyDB for PostgreSQL using your Cloud SQL for PostgreSQL backup. The size limit for these operations is now 15TB. For more information, see Migrate from Cloud SQL for PostgreSQL to AlloyDB.

AlloyDB AI introduces auto embedding generation in Preview. You can now generate large-scale embeddings to leverage in semantic search and Retrieval Augmented Generation (RAG) on text content.

The alloydb_scann extension version 0.1.3 is updated to include the following vector search improvements, which are now Generally Available (GA):

• The columnar engine now automatically includes vector columns in searches, so you don't need to add them to the table manually. For more information, see Perform a vector search.
• You can use the pg_stat_ann_index_creation view for metrics about the number of rows at index creation. For more information, see Vector index metrics.

AlloyDB offers enhanced backups (Preview) that integrate a cluster's backup operations with the Google Cloud Backup and DR Service. This integration provides vaulted backups data with retention lock enforcement, and it uses the Backup and DR Service control plane for centralized policy management and advanced scheduling for AlloyDB clusters. For more information, see Manage enhanced backups.

AlloyDB supports configuring Authorized Networks for Public IP without any CIDR-range restrictions. Use custom organization policies to limit the size and number of Authorized Networks. See Supported custom constraints and operations for examples.

You can now perform time-series forecasting in AlloyDB to predict future trends based on your historical data. This feature supports various forecasting models, including TimesFM, and is available in Preview.

AlloyDB supports the tds_fdw extension, which provides a foreign data wrapper for accessing databases—such as Microsoft SQL Server and Sybase—that use the Tabular Data Stream (TDS) protocol. For more information, see Supported database extensions. This feature is generally available (GA).

You can enable alloydb.enable_cache_aware_costing to turn on cache awareness for AlloyDB for PostgreSQL's query planner. This improves index scan query plans for query performance and reduces IO costs. This feature is in Preview.

The alloydb_scann extension version 0.1.3 is updated to include the following vector search improvements in (Preview):

• You can now automatically create ScaNN indexes that are optimized for search performance or for a balance between index build times and search performance with the auto index feature.
• AlloyDB AI's adaptive filtering for filtered vector searches now dynamically switches between pre-filtering and inline filtering. This enhancement further optimizes query performance by allowing the query optimizer to dynamically choose the most efficient filtering strategy. For more information, see Activate adaptive filtering in AlloyDB AI.
• AlloyDB AI now integrates ScaNN indexes with the columnar engine. You can now accelerate your vector similarity search by loading ScaNN indexes into the columnar engine.
• The alloydb_scann extension now provides a satisfy limit feature that improves query recall for vector searches. If a search returns fewer results than specified in the LIMIT clause, the scan continues until the LIMIT is met or a configured upper bound is reached. To enable this feature, set the scann.satisfy_limit flag to relaxed order. You can also use the scann.max_pct_leaves_to_search flag to configure the upper bound for the search.
• You can enable vector search index recommendations for Scalable Nearest Neighbors (ScaNN) indexes using the AlloyDB index advisor. For more information, see Use the AlloyDB index advisor with query insights or View the index advisor's index recommendations.

• You can configure automatic index maintenance using the following flags:

  • scann.max_background_workers flag to control the number of background workers and increase throughput across multiple indexes.
  • scann.maintenance_background_naptime_s flag to control the minimum delay between maintenance runs.

You can create and manage query plan patches. Query plan patches let you specify the details of the execution plan of your queries. This feature is generally available (GA).

You can now provision, manage, and query your databases using the dedicated Gemini CLI extensions for AlloyDB. The extensions provide full lifecycle control of your database - from provisioning instances to exploring schemas and troubleshooting issues. This feature is available in beta.

Database server compatibility with PostgreSQL version 17 is now generally available (GA). You can create AlloyDB clusters with PostgreSQL 17 compatibility.

The available memory metric now accurately reflects the memory available to AlloyDB by taking into consideration usable memory from the OS page cache. This improvement can lead to a lower value of the metric, which you might notice when you update your version to PG 17 or later. This feature is generally available (GA) and is available for AlloyDB for PostgreSQL version 17 and later. For more information, see System insights metrics reference.

AlloyDB supports C4A Arm VMs on Google's custom-built Axion processors. C4A VMs are available as predefined configurations from 1, 4, 8, 16, 32, 48, 64, and 72 vCPUs, up to 576 GB of DDR5 memory. C4A machines are available in limited regions. For more information, see Considerations when using the C4A Axion-based machine series. This feature is generally available (GA).

Parameterized secure views in AlloyDB for PostgreSQL enhance data security and row access control while using SQL, providing a new secure interface for application developers. Access to this Preview feature no longer requires a signup.

AlloyDB AI natural language delivers secure and accurate responses for application end user natural language questions. Natural language offers fragment-based templates, autogenerated concept types, and SQL summaries. Access to this Preview feature no longer requires a signup.

You can create organization policies with custom constraints for AlloyDB backups and clusters, and a custom constraint with any field for an AlloyDB instance. This feature is generally available (GA).

You can save and manage your SQL scripts in AlloyDB Studio. This feature is in Preview. For more information, see Saved queries overview.

Database server compatibility with PostgreSQL version 17 is now available in Preview. You can create AlloyDB clusters with PostgreSQL 17 compatibility.

You can now create an AlloyDB instance with a specific IP address range using the Google Cloud CLI, Terraform, or REST API. You can also override IP address range allocations configured during cluster creation. For more information, see Create an instance with a specific IP address range. This feature is generally available GA.

You can use the columnar engine to improve the performance of vector similarity searches, specifically K-Nearest Neighbor (KNN) searches, when combined with highly-selective predicate filtering. For more information, see Accelerate your filtered vector search. This feature is in Preview.

You can let AlloyDB automatically create Private Service Connect endpoints for authorized projects when you create Private Service Connect-enabled instances, based on your defined service connection policy. This feature is generally available (GA).

You can create AlloyDB clusters with Private Services Connect through the Google Cloud console.

You can now start, stop, and restart your primary and read pool AlloyDB instances using the Google Cloud console and the Google Cloud CLI. This feature is generally available (GA).

AlloyDB for PostgreSQL supports the pg_ivm extension, which provides incremental view maintenance for materialized views.

AlloyDB AI query engine (Preview) lets you combine natural language with SQL using operators like ai.if, ai.rank, and ai.generate. You can generate embeddings on images, videos, and text using multimodal embedding models, and you can use Google DeepMind's latest text embedding model into model endpoint management.

The default major version of PostgreSQL compatibility for new AlloyDB clusters is now PostgreSQL 16.

Due to a change to report replay_lsn more accurately during parallel replay, metrics might show a slightly higher replication lag.

AlloyDB supports IAM authentication in AlloyDB Studio. For more information, see Choose a database authentication method.

You can migrate from Cloud SQL for PostgreSQL to AlloyDB for PostgreSQL using your Cloud SQL for PostgreSQL backup (GA). The Google Cloud CLI is also supported. For more information, see Migrate from Cloud SQL for PostgreSQL to AlloyDB.

AlloyDB lets you configure a deny maintenance period on clusters running the latest version. The feature is generally available (GA).

You can now build a vector embedding Extract, Transform, Load (ETL) pipeline that lets you generate and ingest embeddings from files or real time sources to AlloyDB using Google Cloud Dataflow. For more information, see Build realtime vector embedding pipeline for AlloyDB with Dataflow.

8 AlloyDB recommenders are now generally available (GA). For more information, see the following pages:

• Enable backup plans
• Enable database auditing
• Enable high availability
• Enable password policy
• Enforce SSL/TLS encryption
• Increase backup retention
• Increase cluster storage quota
• Optimize underprovisioned instances

AlloyDB now supports managed connection pooling in Preview. You can use managed connection pooling on your instances to improve the reliability, scalability, and performance of your workloads by optimizing resource utilization. For more information, see Configure managed connection pooling.

AlloyDB for PostgreSQL supports a 1 virtual central processing unit (vCPU) configuration with 8GB of memory, which is suitable for development and sandbox environments. For information about 1 vCPU supported regions and limitations, see Considerations when using 1 vCPU. This feature is in Preview.

AlloyDB supports C4A Arm VMs on Google's custom-built Axiom processors. C4A VMs are available as predefined configurations from 1, 4, 8, 16, 32, 48, 64, and 72 vCPUs, up to 576 GB of DDR5 memory. C4A machines are available in limited regions. This feature is in Preview. For more information, see Considerations when using the 1 vCPU machine type.

AlloyDB supports AI-assisted troubleshooting that helps you resolve complex database performance issues like slow queries and high load. AI-assisted troubleshooting is available in Preview.

AlloyDB for PostgreSQL supports parameterized secure views, which provide a secure interface for application developers by improving data security and row access control while using SQL. This feature is in (Preview). For more information, see Parameterized secure views overview.

The alloydb_scann extension is updated to include the following vector search improvements. These features are generally available (GA):

• Inline filtering enables the execution of vector search and filter evaluation through the combined use of vector and secondary indexes. For more information, see "Inline filtering" in the documentation for AlloyDB PostgreSQL and AlloyDB Omni 15.7.1 and 16.3.0.

• You can let AlloyDB automatically create multiple parallel workers during index creation when the dataset grows, leading to faster build times. For more information, see "Build indexes in parallel" in the documentation for AlloyDB PostgreSQL and AlloyDB Omni 15.7.1 and 16.3.0.

• A distribution histogram is available in the pg_stat_ann_indexes view, which helps you understand the distribution of vectors between partitions of your ScaNN index. For more information, including recommendations about tuning the distributionpercentile metric, see "Tuning metrics" in the documentation for AlloyDB PostgreSQL, and AlloyDB Omni 15.7.1 and 16.3.0.

• You can use a query recall evaluator to find the recall for a vector query for a given configuration, and to tune your parameters to achieve the desired vector query recall results for different vector indexes. For more information, see "Measure vector query recall" in the documentation for AlloyDB PostgreSQL, and AlloyDB Omni 15.7.1 and 16.3.0.

The alloydb_scann extension is updated to include the following vector search improvements in (Preview):

• You can enable auto-maintenance for your ScaNN index and let incrementally manage the index such that when your dataset grows, AlloyDB splits large outlier partitions, and tries to provide better QPS and search results. For more information, see "Maintain indexes automatically" in the documentation for AlloyDB PostgreSQL and AlloyDB Omni 15.7.1 and 16.3.0.

• Adaptive filtering for ScaNN significantly improves the speed of filtered vector searches. Adaptive filtering automatically selects the most efficient filtering method at runtime. For more information, see "Filtered vector search" and "Adaptive filtering" in the documentation for AlloyDB for PostgreSQL and AlloyDB Omni 15.7.1 and 16.3.0.

• You can enable index auto maintenance and adaptive inline filtering together using the scann.enable_preview_features Grand Unified Configuration (GUC) parameters. For more information, see "AlloyDB flags" for AlloyDB for PostgreSQL and AlloyDB Omni 15.7.1 and 16.3.0.

AlloyDB AI query engine that builds on model endpoint management, and adds support for AI operators and Vertex AI multimodal and ranking models is available in (Preview). You can combine natural language phrases with SQL queries, like ai.if() for filters and joins, ai.rank() for ordering using ranking models, and ai.generate() for generating summaries of your data, and generate multimodal embeddings.

AlloyDB AI natural language (Preview) delivers secure and accurate responses for application end user natural language questions. For more information, see AlloyDB AI natural language overview.

When the ScaNN index creation updates the reltuples statistics of a heap table, performance might be degraded for queries involving that table. For information to mitigate the issue, see "Analyze your indexed table" in the documentation for AlloyDB for PostgreSQL and AlloyDB Omni.

If your cluster is encrypted with a customer-managed encryption key (CMEK), and no specific CMEK key is configured for continuous or automated backups, then backups will be created with the cluster CMEK. For more information, see About CMEK and Configure backup plans.

Performing an in-place major version upgrade of your AlloyDB cluster is generally available (GA). You can upgrade your AlloyDB cluster to any higher supported PostgreSQL version. For information about supported PostgreSQL versions, see Database version policies.

You can use a query recall evaluator (Preview) to find the recall for a vector query for a given configuration, and to tune your parameters to achieve the desired vector query recall results. For more information, see Measure vector query recall.

You can enforce specific tags on AlloyDB cluster and backup resources using custom organization policies. If a mandatory tag is missing or does not have a value set, AlloyDB resource creation fails. This feature is available in Preview.

Outbound connectivity for Private Service Connect-enabled AlloyDB clusters is generally available (GA). Enabling outbound connectivity allows secure connection between your project and an AlloyDB instance during outbound operations such as migrations or foreign data wrappers (FDW).

You can automatically create CMEKs using Cloud KMS Autokey to protect your AlloyDB resources. This feature is generally available (GA).

AlloyDB for PostgreSQL is available in the following region: europe-north2 (Stockholm). For more information, see AlloyDB Locations.

AlloyDB's cross-region replication supports up to five secondary regions. You can use additional secondary regions to further harden disaster recovery response, or to serve geographically distributed workloads. For more information, see Cross-region replication overview and Work with cross-region replication.

The alloydb_scann extension is updated to include the following vector search improvements in Preview:

• AlloyDB for PostgreSQL introduces inline filtering for vector search. With inline filtering, SQL filter evaluation is performed at the same time as vector search. This feature mitigates potential issues from existing pre and post-filter evaluation mechanisms. For more information about scann.enable_inline_filtering, see ScaNN index reference.
• A distribution histogram is available in the pg_stat_ann_indexes view, which helps you understand the distribution of vectors between partitions and num_leaves of your ScaNN index. For more information, including recommendations about tuning the distributionpercentile metric, see Tuning metrics.

You can use an API to import data to AlloyDB for PostgreSQL clusters. This feature is generally available (GA). You can also cancel the import of data and check the status of the import. For more information, see Import a CSV file and Import a SQL file.

You can perform an in-place major version upgrade of your AlloyDB cluster to PostgreSQL version 16 from PostgreSQL version 14 or 15. This feature is available in Preview.

Support for advanced query insights, index advisor, and active queries is now generally available (GA) in AlloyDB for PostgreSQL.

You cannot enable advanced query insights on clusters with secondary instances. See Limitations for more information.

The extension vector, which includes pgvector functions and operators, is updated to version 0.8.0.

The AlloyDB Omni Kubernetes operator version 1.3.0 is generally available (GA). Version 1.3.0 includes the following new features and changes:

• The Kubernetes operator 1.3.0 supports connection pooling (Preview).

• You can put the database in maintenance mode to make a maintenance update or repair a pod.

• You can create replication slots and users for logical replication via the Operator API from your database instance to subscribed applications.

• This release of the Kubernetes operator adds support for kube-state-metrics so that you can use Prometheus or a Prometheus-compatible scraper to consume and display custom resource metrics like DBCluster Backup, and DBInstance. For more information, see Monitor AlloyDB Omni Kubernetes operator custom resources.

• When you create a new database cluster, this version of the Kubernetes operator creates read-only (RO) and read-write (RW) load balancers concurrently, which reduces the time that it takes for the database cluster to be ready for connections and queries.

• Configurable log rotation has a default retention time of seven days, and each archived file is individually compressed using Gzip. For more information, see Configure AlloyDB Omni log rotation.

• Various bug fixes and performance improvements.

AlloyDB lets you monitor the following additional monitoring metrics through the Cloud Monitoring dashboard. These metrics are available in Preview.

• The instance/postgres/ultrafastcache_hitrate and node/postgres/ultrafastcache_hitrate metrics help in identifying any performance issue due to caching on instances or individual nodes.
• The node/postgres/backends_by_state, node/postgres/backends, node/postgres/wait_count, and node/postgres/wait_time metrics help in tracking node health.

You can use an API to export data from AlloyDB clusters. This feature is generally available (GA). You can also cancel the export of data. For more information, see Export a CSV file and Export a SQL file.

AlloyDB System insights offers a unified, customizable database monitoring dashboard that includes predefined metrics and other Google Cloud metrics. This feature is generally available (GA). For more information, see Create a custom dashboard.

The Perform a vector search tutorial describes how to set up and perform a vector search in AlloyDB for PostgreSQL. You can learn how to perform K-nearest neighbor (KNN) and approximate nearest-neighbor (ANN) with a ScaNN vector index.

The AlloyDB database performance snapshot reports feature is generally available (GA).This feature lets you improve your database performance by using a report that compares snapshots of system metrics between two different points in time. For more information, see Optimize database performance by comparing performance snapshots.

You can set up AlloyDB clusters using a copy of your Cloud SQL for PostgreSQL backup. This feature is in Preview. For more information, see Migrate from Cloud SQL for PostgreSQL to AlloyDB.

Model endpoint management is generally available (GA) for both AlloyDB and AlloyDB Omni. You can use sample templates to register model endpoints. For more information, see Register and call remote AI models in AlloyDB or Register and call remote AI models in AlloyDB Omni.

AlloyDB for PostgreSQL is now available in the following region: northamerica-south1 (Mexico). For more information, see AlloyDB Locations.

AlloyDB free trial clusters are now available in all regions. For more information, see the AlloyDB free trial clusters overview.

The extension vector, which includes pgvector functions and operators, is updated to version 0.7.4.

If you are dropping an AlloyDB database that is larger than 64 TiB, then any write operations on other AlloyDB databases are paused until the drop operation is completed.

AlloyDB now supports up to 128 TiB storage per cluster in all regions where AlloyDB is available.

The extension pg_ivm version 1.9 has been added to extensions supported by AlloyDB Omni.

The following extensions are updated:

• google_ml_integration from 1.3 to 1.4.2
• pg_partman from 4.7.4 to 5.0.1
• pglogical from 2.4.4 to 2.4.5
• pgtt from 3.0.0 to 4.0.0
• vector is updated from 0.7.0 to 0.7.4

In AlloyDB Omni versions 15.5.5 and earlier, parameterized view features were available in the alloydb_ai_nl extension. Starting in version 15.7.0, parameterized view features are available in the parameterized_views extension, which you must create before you use parameterized views. Also starting in version 15.7.0, the related function, google_exec_param_query, has been renamed to execute_parameterized_query and is available in the